GPUStack CVE-2026-102895 is on CISA's KEV list: the patch target, the forensic hunt and a hardening checklist for self-hosted inference
Blog post from Qovery
GPUStack CVE-2026-102895 is a critical, unauthenticated administrator bypass affecting versions earlier than 2.2.0: requests that appear to originate from 127.0.0.1 could gain administrator access, including through a same-host reverse proxy or a forged IP header. CISA added the vulnerability to its Known Exploited Vulnerabilities catalog on 9 November 2026, set a 12 November remediation deadline, and flagged the need for forensic triage after exploitation was observed in the wild. The recommended target is GPUStack 2.3.1, since 2.2.0 fixes the bypass but not a separate SAML authentication flaw addressed in 2.2.2 or later advisories fixed in subsequent releases. Teams are advised to identify all deployments and their exposure, restrict vulnerable servers before upgrading, preserve logs and other evidence, investigate for unauthorized administrator activity, persistent API keys, custom backends and worker-host shells, and rotate credentials only after patching. Longer-term protections include keeping management interfaces on private networks, using an authenticated gateway that strips untrusted client-IP headers, and applying Kubernetes network restrictions, egress controls, admission policies, image pinning and workload identities. The guidance also notes that vLLM’s built-in API-key protection does not cover every endpoint, reinforcing the need for gateway-level authentication and path restrictions across self-hosted inference services.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Kubernetes | 24 | 4,377 | 419 | 114 | -15% |
| MCP | 6 | 9,757 | 959 | 203 | -18% |
| Secrets Management | 5 | 2,876 | 539 | 148 | -7% |
| AI Agents | 3 | 6,233 | 1,230 | 245 | -11% |
| Local AI | 3 | 94 | 59 | 19 | -44% |
| LLM | 2 | 3,674 | 823 | 199 | -44% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.